Papers
Topics
Authors
Recent
Search
2000 character limit reached

Ski Rental with Distributional Predictions of Unknown Quality

Published 24 Feb 2026 in cs.LG and cs.DS | (2602.21104v1)

Abstract: We revisit the central online problem of ski rental in the "algorithms with predictions" framework from the point of view of distributional predictions. Ski rental was one of the first problems to be studied with predictions, where a natural prediction is simply the number of ski days. But it is both more natural and potentially more powerful to think of a prediction as a distribution p-hat over the ski days. If the true number of ski days is drawn from some true (but unknown) distribution p, then we show as our main result that there is an algorithm with expected cost at most OPT + O(min(max({eta}, 1) * sqrt(b), b log b)), where OPT is the expected cost of the optimal policy for the true distribution p, b is the cost of buying, and {eta} is the Earth Mover's (Wasserstein-1) distance between p and p-hat. Note that when {eta} < o(sqrt(b)) this gives additive loss less than b (the trivial bound), and when {eta} is arbitrarily large (corresponding to an extremely inaccurate prediction) we still do not pay more than O(b log b) additive loss. An implication of these bounds is that our algorithm has consistency O(sqrt(b)) (additive loss when the prediction error is 0) and robustness O(b log b) (additive loss when the prediction error is arbitrarily large). Moreover, we do not need to assume that we know (or have any bound on) the prediction error {eta}, in contrast with previous work in robust optimization which assumes that we know this error. We complement this upper bound with a variety of lower bounds showing that it is essentially tight: not only can the consistency/robustness tradeoff not be improved, but our particular loss function cannot be meaningfully improved.

Authors (2)

Summary

  • The paper introduces an algorithm that needs no knowledge of prediction error and achieves additive loss O(min(√b·max(EMD,1), b log b)), combining O(√b) consistency with O(b log b) robustness.
  • The method delays the predicted optimal purchase time and truncates overly long rentals using predicted-tail structure, while a λ-interpolated variant offers a tunable balance between prediction-dependent performance and classical competitiveness.
  • The paper proves matching lower bounds showing that linear EMD dependence is impossible, the √b·max(EMD,1) scaling is tight within its natural form, and O(√b) consistency necessarily entails Ω(b log b) robustness.

Problem setting and motivation

This paper studies the classical ski rental problem within the algorithms-with-predictions framework, replacing the standard point prediction with a distributional prediction. The setup is as follows: rental costs 1 per day, buying costs b>1b > 1, and the true number of ski days TT is drawn from an unknown distribution pp over [N][N]. The algorithm receives a predicted distribution p^\hat{p}, measured against pp via Earth Mover's Distance (EMDEMD, i.e., Wasserstein-1). A deterministic policy AKA_K rents for KK days and buys on day K+1K+1; the benchmark is the expected cost of the optimal policy for the true distribution, TT0, and additive loss is TT1.

Two framing choices matter. First, benchmarking against TT2 rather than the optimum in hindsight is necessary: for a bimodal truth placing mass TT3 at TT4 and TT5 at TT6, any policy incurs TT7 loss relative to hindsight even when TT8 is known exactly. Second, collapsing TT9 to a point (via sampling or MLE) can be arbitrarily lossy for multimodal predictions, motivating a genuinely distributional treatment.

The key distinction from prior work concerns what the algorithm knows about prediction quality. Besbes, Ma, and Mouchtaki assume pp0 is known to the algorithm; this enables stronger bounds of pp1 but makes robustness trivial (one can always fall back to buying immediately for worst-case loss pp2). This paper removes that assumption — answering an open question posed explicitly by Besbes et al. — which forces robustness to become a genuine design constraint.

Main upper bound

The central result is an algorithm achieving, without any knowledge of pp3,

pp4

This yields consistency pp5 (additive loss at zero error) and robustness pp6 (additive loss at unbounded error), simultaneously and without knowing pp7. Notably, whenever pp8 the loss is strictly sublinear in pp9, beating the trivial bound.

The algorithm has two components:

  1. Base algorithm: compute the optimal policy [N][N]0 for [N][N]1, then delay purchase by [N][N]2 additional days. The analysis compares [N][N]3 to [N][N]4 through the optimal transport plan between [N][N]5 and [N][N]6, classifying transport maps as "good" (self-bounding by their [N][N]7 contribution) or "bad" (split into short-distance moves, controlled by structural properties of optimal policies in [N][N]8, and long-distance moves, whose total mass is at most [N][N]9).
  2. Truncation step for robustness: let p^\hat{p}0 be the first day where the tail mass of p^\hat{p}1 beyond p^\hat{p}2 is at most p^\hat{p}3; if p^\hat{p}4, buy at day p^\hat{p}5 instead. The truncation loss analysis uses a clean tail lemma: for any p^\hat{p}6 and p^\hat{p}7, p^\hat{p}8, so the probability that truncation triggers under the truth is at most p^\hat{p}9, bounding extra cost by pp0.

Robustness rests on two structural lemmas about the predicted tails pp1: optimality of pp2 forces the predicted tail to decay geometrically before pp3 (with ratio pp4 per step backwards), and dyadic tail levels must be crossed within windows of length pp5. Together these imply the truncation time is always pp6 regardless of how misleading pp7 is. The authors note the base-algorithm bound also follows from equation (E-28) of Besbes et al.'s appendix, though their proof was developed independently and is arguably more direct.

Further robustification

For applications demanding worst-case guarantees below pp8, the paper gives a pp9-interpolated algorithm, following Purohit et al.: it clamps the base algorithm's buying time toward day EMDEMD0 from either side, buying at day EMDEMD1 or EMDEMD2 as appropriate. It satisfies

EMDEMD3

where EMDEMD4. At EMDEMD5 this recovers the main theorem; at EMDEMD6 it recovers the classical factor-EMDEMD7 competitive ratio. This mirrors the point-prediction tradeoff of Purohit et al., generalized to distributions.

Matching lower bounds

The paper supplies four lower bounds, all holding against randomized algorithms (via Yao's lemma), establishing near-optimality of both the bound's shape and the consistency–robustness tradeoff.

Result Statement
No linear-in-EMD generalization No algorithm achieves loss EMDEMD8 for any EMDEMD9
Necessity of AKA_K0 No algorithm achieves loss AKA_K1
Tightness of AKA_K2 No algorithm achieves loss AKA_K3
Pareto-optimal tradeoff Consistency AKA_K4 forces robustness AKA_K5

Several points deserve emphasis. First, the impossibility of AKA_K6 means the point-prediction guarantee of Purohit et al. does not extend to distributions: error and AKA_K7 must combine nonlinearly. Second, the third lower bound matches the main upper bound exactly within the natural family AKA_K8, so the non-robust component of the bound cannot be improved in this form. Third, the Pareto-optimality result is proved via an explicit adversarial prediction whose tail decays at rate AKA_K9 up to KK0 (making KK1 optimal) and more slowly thereafter; any algorithm consistent on this prediction must rent until KK2, suffering KK3 loss when the truth is a point mass at KK4.

There are caveats. The tightness of KK5 holds only within the restricted functional form KK6; the paper does not rule out better bounds of a different shape. Also, in the limited setting where the truth is a point distribution, the linear-in-EMD lower bound does not apply: the paper gives a deterministic algorithm achieving loss KK7, derandomizing the sampling-based approach.

Relative to the three closest papers, the contribution is well delineated. Purohit, Svitkina, and Kumar supply the point-prediction template (including the KK8-interpolation technique reused here). Diakonikolas et al. address sample complexity of learning the true distribution within structured families and provide no error-dependent guarantees. Besbes et al. study essentially the same problem under known KK9 via robust optimization; their stronger K+1K+10 bound exploits knowledge of K+1K+11, and their explicit open question — performance without knowledge of K+1K+12 — is precisely what this paper resolves. The work also connects to binary search with distributional predictions, which posed a similar question in a different problem.

Limitations and open questions

The paper concedes several boundaries. The robustness term K+1K+13 carries an extra logarithmic factor over the trivial worst-case loss of K+1K+14 achievable when K+1K+15 is known; whether unknown-quality robustness below K+1K+16 is possible while retaining sublinear consistency remains unresolved except through the K+1K+17-interpolation, which degrades the low-error regime. The matching lower bound for K+1K+18 is conditional on the multiplicative form of the bound, so a qualitatively different (e.g., additive-plus-multiplicative) improvement is not excluded. Finally, the analysis assumes deterministic policies for the upper bounds and finite support K+1K+19, and the benchmark choice excludes comparisons to the optimum in hindsight, where TT00 loss is unavoidable.

Conclusion

This paper settles the ski rental problem with distributional predictions of unknown quality: an TT01-consistent, TT02-robust algorithm exists without knowledge of the prediction error, and its consistency–robustness tradeoff is Pareto-optimal, with the loss function itself essentially tight among natural forms. The results close an open question of Besbes et al. and establish that the transition from point to distributional predictions genuinely changes the achievable landscape — linear dependence on prediction error is impossible, and the correct scaling is TT03 capped at TT04.

Paper to Video (Beta)

No one has generated a video about this paper yet.

Whiteboard

No one has generated a whiteboard explanation for this paper yet.

Open Problems

We haven't generated a list of open problems mentioned in this paper yet.

Tweets

Sign up for free to view the 1 tweet with 13 likes about this paper.